We are a compensation and performance management product company, specializing in delivering better results than the typical Talent Management vendors with our focus on results. For the past year, we have been exploring using machine learning and predictive analytics. Our clients have traditionally been HR administrators and the line managers of their respective organizations. We started with different programs like predicting next year’s salary budgets for our customer organization, including doing sentiment analysis. The first break-through came through a tool we created to detect anomalous data, catching data errors before releasing to managers. Since then we have found additional uses for our Predictive Analytics tool, which are having a real impact on results.
Examples of the promise of machine learning include:
1. Anomaly Detection
Anomaly Detection – through our flagship compensation and performance management product suite – Compass Rewards – provides an extremely useful method for cleaning up bad HR data. The source of the HR data is very diverse – from core HR and Payroll system representing the critical employee data elements. Despite the customer’s best efforts to sanitize the data before feeding it into our system, discrepancies do creep in. Usually, these discrepancies are caught toward the end of the compensation or performance cycles, which makes it expensive to correct. We created an anomaly detection tool based on machine learning algorithms like linear regression and KNN. The Anomaly Detection tool identifies clear outliers which fall outside a standard deviation for a given employee information data set. Once the outlier is identified, our team samples it to see if it is indeed an anomaly, then the results are sent to the customer for their action. The Anomaly Detection tool has proved so effective that it has become useful to not only audit the compensation and performance cycles, but for correcting the records in upstream HRM systems.
2. Supervised Learning Algorithms
We used supervised learning algorithms and detected interesting patterns on one of our customer’s employee data sets following a completed compensation cycle. We found that the higher the employee’s age, the lower the merit increase they received, even when compared to similar employees regardless of age. Some of the older employees received some of the lowest raises for doing the same job at the same level of achievement. Job descriptions, the type of work and the number of hours of worked were all the same when comparing the older employees to their younger peers. This highlighted a bias toward younger workers that discounted the value of experience and enabled HR leaders to focus on ensuring equity for high performing employees, regardless of age. Other examples include generating gender-diversity reports based on similar algorithms.
We are now in an expansion phase, where we continue to invest and work on predictive analytics and machine learning aimed at providing HR leaders and executives actionable data. Will keep you posted – stay tuned.